func checkMasterRole(node *topo.Node, ismaster bool) (bool, error) {
addr := node.Addr()
info, err := redis.FetchInfo(addr, "replication")
if err != nil {
return false, fmt.Errorf("Connect %s failed", addr)
}
if info.Get("role") == "master" {
return true, nil
} else {
return false, nil
}
}
func checkSlaveRepliStatusOk(node *topo.Node) (bool, error) {
addr := node.Addr()
info, err := redis.FetchInfo(addr, "all")
if err != nil {
return false, err
}
if info.Get("role") == "master" {
return false, nil
}
if info.Get("master_link_status") == "down" {
return false, nil
} else if info.Get("loading") == "1" {
return false, nil
} else {
return true, nil
}
}
// 失败返回-1
func fetchReplOffset(addr string) int64 {
info, err := redis.FetchInfo(addr, "Replication")
if err != nil {
return -1
}
if info.Get("role") == "master" {
offset, err := info.GetInt64("master_repl_offset")
if err != nil {
return -1
} else {
return offset
}
}
offset, err := info.GetInt64("slave_repl_offset")
if err != nil {
return -1
}
return offset
}
func (cs *ClusterState) RunFailoverTask(oldMasterId, newMasterId string) {
new := cs.FindNodeState(newMasterId)
old := cs.FindNodeState(oldMasterId)
if old == nil {
log.Warningf(oldMasterId, "Can't run failover task, the old dead master lost")
return
}
if new == nil {
log.Warningf(oldMasterId, "Can't run failover task, new master lost (%s)", newMasterId)
old.AdvanceFSM(cs, CMD_FAILOVER_END_SIGNAL)
return
}
// 通过新主广播消息
redis.DisableRead(new.Addr(), old.Id())
redis.DisableWrite(new.Addr(), old.Id())
c := make(chan error, 1)
go func() {
//choose failover force or takeover in case of arbiter
cluster := cs.cluster
rs := cluster.FindReplicaSetByNode(old.Id())
if cluster.HasArbiter() || cluster.IsClusterDown() {
//use failover takeover
c <- redis.SetAsMasterWaitSyncDone(new.Addr(), true, true, rs)
} else {
//use failover force
c <- redis.SetAsMasterWaitSyncDone(new.Addr(), true, false, rs)
}
}()
select {
case err := <-c:
if err != nil {
log.Eventf(old.Addr(), "Failover request done with error(%v).", err)
} else {
log.Eventf(old.Addr(), "Failover request done, new master %s(%s).", new.Id(), new.Addr())
}
case <-time.After(20 * time.Minute):
log.Eventf(old.Addr(), "Failover timedout, new master %s(%s)", new.Id(), new.Addr())
}
// 重新读取一次，因为可能已经更新了
roleChanged := false
node := cs.FindNode(newMasterId)
if node.IsMaster() {
roleChanged = true
} else {
for i := 0; i < 10; i++ {
info, err := redis.FetchInfo(node.Addr(), "Replication")
if err == nil && info.Get("role") == "master" {
roleChanged = true
break
}
log.Warningf(old.Addr(),
"Role of new master %s(%s) has not yet changed, will check 5 seconds later.",
new.Id(), new.Addr())
time.Sleep(5 * time.Second)
}
}
if roleChanged {
log.Eventf(old.Addr(), "New master %s(%s) role change success", node.Id, node.Addr())
// 处理迁移过程中的异常问题，将故障节点（旧主）的slots转移到新主上
oldNode := cs.FindNode(oldMasterId)
if oldNode != nil && oldNode.Fail && oldNode.IsMaster() && len(oldNode.Ranges) != 0 {
log.Warningf(old.Addr(),
"Some node carries slots info(%v) about the old master, waiting for MigrateManager to fix it.",
oldNode.Ranges)
} else {
log.Info(old.Addr(), "Good, no slot need to be fix after failover.")
}
} else {
log.Warningf(old.Addr(), "Failover failed, please check cluster state.")
log.Warningf(old.Addr(), "The dead master will goto OFFLINE state and then goto WAIT_FAILOVER_BEGIN state to try failover again.")
}
old.AdvanceFSM(cs, CMD_FAILOVER_END_SIGNAL)
// 打开新主的写入，因为给slave加Write没有效果
// 所以即便Failover失败，也不会产生错误
redis.EnableWrite(new.Addr(), new.Id())
}
